Some days ago I started a little bash script that should sum up the number of pages and file size of all PDF's in a folder. It's working quite well now but there's still one thing I don't understand.
Why is sed always failing if shopt -s nullglob is set? Does somebody know why this happens?
I'm working with GNU Bash 4.3 and sed 4.2.2 in Ubuntu 14.04.
set -u
set -e
folder=$1
overallfilesize=0
overallpages=0
numberoffiles=0
#If glob fails nothing should be returned
shopt -s nullglob
for file in $folder/*.pdf
do
# Disable empty string if glob fails
# (Necessary because otherwise sed fails ?:|)
#shopt -u nullglob
# This command is allowed to fail
set +e
pdfinfo="$(pdfinfo "$file" 2> /dev/null)"
ret=$?
set -e
if [[ $ret -eq 0 ]]
then
#Remove every non digit in the result
sedstring='s/[^0-9]//g'
filesize=$(echo -e "$pdfinfo" | grep -m 1 "File size:" | sed $sedstring)
pages=$(echo -e "$pdfinfo" | grep -m 1 "Pages:" | sed $sedstring)
overallfilesize=$(($overallfilesize + $filesize))
overallpages=$(($overallpages+$pages))
numberoffiles=$(($numberoffiles+1))
fi
done
echo -e "Processed files: $numberoffiles"
echo -e "Pagesum: $overallpages"
echo -e "Filesizesum [Bytes]: $overallfilesize"
Here's a simpler test case for reproducing your problem:
#!/bin/bash
shopt -s nullglob
pattern='s/[^0-9]//g'
sed $pattern <<< foo42
Expected output:
42
Actual output:
Usage: sed [OPTION]... {script-only-if-no-other-script} [input-file]...
(sed usage follows)
This happens because s/[^0-9]//g is a valid glob (matching a dir structure like like s/c/g), and you asked bash to interpret it. Since you don't have a matching file, nullglob kicks in and removes the pattern entirely.
Double quoting prevents word splitting and glob interpretation, which is almost always what you want:
#!/bin/bash
shopt -s nullglob
pattern='s/[^0-9]//g'
sed "$pattern" <<< foo42
This produces the expected output.
You should always double quote all your variable references, unless you have a specific reason not to.